Research on DSC-MB-PSPNet Semantic Segmentation inStreet-scene Autonomous Driving

It presented a lightweight real-time semantic segmentation model for city autonomous driving. A deep separable convolution, multi-branch and Pyramid scene parsing network fusion structure (DSC-MB-PSPNet) was proposed to ensure the model has better characterization ability and real-time operation, an...

Full description

Saved in:
Bibliographic Details
Main Authors: HU Yunqing, PAN Wenbo, HOU Zhichao, JIN Weizheng, YU Huan
Format: Article
Language:zho
Published: Editorial Office of Control and Information Technology 2020-01-01
Series:Kongzhi Yu Xinxi Jishu
Subjects:
Online Access:http://ctet.csrzic.com/thesisDetails#10.13889/j.issn.2096-5427.2020.04.100
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:It presented a lightweight real-time semantic segmentation model for city autonomous driving. A deep separable convolution, multi-branch and Pyramid scene parsing network fusion structure (DSC-MB-PSPNet) was proposed to ensure the model has better characterization ability and real-time operation, and inhibitory cross entropy loss function was presented to reduce the impact of sample imbalance and eliminate the serious imbalance of pixel number among samples in street scenes autonomous driving. At the same time, in order to improve the training efficiency and make the model easier to converge, the multi-stage loss function calculation method was added in the model. DSC-MB-PSPNet was evaluated on the Cityscapes and self-built datasets. The results show that under the condition of large resolution input, the proposed model achieves state-of-the-art performance and can work in real-time.
ISSN:2096-5427